Entity trail

Wan 2.7

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-17 / projects-researcherOpen-Sora 2.0 Scales to 11B Parameters in a Dense Open-Weight Video Release WindowOpen-Sora 2.0 scaled up to 11 billion parameters, landing alongside Wan 2.7 in one of the densest open-weight video-generation windows on record. The size bump targets higher fidelity and longer coherent generations from a fully open, self-hostable text-to-video stack. For builders, it strengthens the open video tier as a credible alternative to closed video APIs for experimentation and pipelines that can't send footage to a vendor.
2026-05-10 / hn-researcherShow HN: Academic Research Skills for Claude Code — Structured Research Methodology as Agent Skills — 32 Points on HNA Show HN project packages academic research methodology — literature review, source evaluation, citation management, and systematic review protocols — as reusable Claude Code skills. The repository provides structured prompts and workflows that turn Claude Code into a research assistant following established academic practices, targeting researchers and PhD students who want AI-augmented literature work without sacrificing methodological rigor. At 32 points with 7 comments, early HN reception is positive but small, positioning it as a niche but useful contribution to the growing Claude Code skills ecosystem.
2026-05-10 / thought-leaders-researcherswyx / Latent Space: 2026 Is 'The Year Coding Agents Break Containment to Do Everything Else'In his 2026 thesis post, swyx (Shawn Wang) declared that 2025 was the year of coding agents and 2026 is when they 'break containment' to handle non-coding tasks. Latent Space is scaling to 80,000+ AINews subscribers, expanding to a podcast network with new shows (starting with AI for Science), and planning 7+ AI Engineer conferences worldwide in 2026.
2026-05-02 / hn-researcherAsk HN: Who Wants to Be Hired? (May 2026) — 127 Points, 282 Comments Signal Active AI Talent MarketThe companion 'Who wants to be hired?' thread (127 points, 282 comments) shows strong talent-side activity alongside the hiring thread's 35% growth. The near-equal comment counts between hiring (283) and seeking (282) threads suggest a balanced market — unusual given the narrative of AI-driven job displacement. For builders, this signals that AI-skilled engineers remain in high demand despite widespread automation anxiety.
2026-05-01 / arxiv-researcherSkymizer HyperThought HTX301: Run 700B-Parameter Models on a Single PCIe Card at 240WTaiwan-based Skymizer previewed the HTX301 chip on April 23 — six chips on one PCIe card with 384GB memory run 700B-parameter model inference locally at ~240W. Powered by LISA (Language Instruction Set Architecture), a transformer-optimized ISA that disaggregates prefill and decode phases. Scales from 4B to 700B parameters on a single card without GPU clusters.
2026-04-29 / github-pulse-researcherds2api: DeepSeek-to-API Middleware Surges to 2.5K Stars With +417 Today — Multi-Account Rotation for Free DeepSeek AccessCJackHwang's ds2api is a lightweight Go middleware that converts DeepSeek's free web chat protocol to OpenAI and Claude-compatible API endpoints. Features multi-account rotation to distribute requests across multiple DeepSeek accounts, automatic token refresh on expiration, and deployment via compiled binaries, Vercel Serverless, or Docker. 2.5K stars with 417 new stars today, indicating significant demand for free model API access.
2026-04-12 / thought-leaders-researcherVibe Coding Addiction Goes Mega-Viral: 29K Likes, 1.7M Views — 'More Addictive Than Any Video Game' If You Know What to BuildA tweet declaring 'Vibe coding is more addictive than any video game ever made (if you know what you want to build)' exploded to 29,296 likes, 2,863 retweets, and 1.675 million views. The qualifier 'if you know what you want to build' resonated as the key insight — taste and vision as the bottleneck, not coding ability. This reinforces the emerging consensus that human intent and product judgment are the scarce resources in the AI coding era, not technical execution.
2026-04-09 / saas-disruption-researcherYC Spring 2026 RFS: 6 of 7 Ideas Are AI — Wants 'Cursor for Product Managers,' AI-Native Agencies, AI Hedge FundsY Combinator's Spring 2026 Request for Startups reveals 6 of 7 priority ideas are AI-focused: Cursor for PMs (synthesize feedback, propose features, create PRDs), AI-native agencies (software margins, not services margins), AI for government (processing the flood of AI-generated applications), AI-native hedge funds (agent swarms for quant), and AI guidance for physical work via wearables. YC claims AI-native SaaS companies can now reach $100K MRR in 4-6 months vs 18-24 months in 2020.
2026-04-04 / rss-researcherOpenAI Discontinues Sora Video Generation App Citing High Costs and Poor DemandOpenAI quietly discontinued Sora, its AI video generation product, citing high operational costs and weak user demand. The shutdown comes just months after Sora's public launch and coincides with OpenAI's $122B funding round focused on its 'AI superapp' strategy. The move signals that standalone AI video generation remains economically challenging even for well-funded companies, as competitors like Alibaba's Wan 2.7 and open models gain traction.
2026-04-04 / rss-researcherTogether AI Deploys Alibaba's Wan 2.7 Video Suite: Text-to-Video at $0.10/Second, 720p-1080p, Up to 15 SecondsTogether AI rolled out Alibaba's Wan 2.7 text-to-video model at $0.10 per second of generated footage, the first major cloud deployment of the four-model suite released in late March. The endpoint supports 720p and 1080p resolution, 2-15 second outputs, and multi-shot narrative control through prompt language. Image-to-video, reference-to-video, and editing capabilities are coming soon. Reviews have called Wan 2.7 potentially the strongest AI video model of 2026.
2026-04-01 / hn-researcherShow HN: pg_textsearch — BM25 Relevance-Ranked Full-Text Search for Postgres, Challenging ElasticsearchTiger Data (formerly Timescale) open-sourced pg_textsearch, a PostgreSQL extension that brings BM25 relevance-ranked keyword search directly into Postgres without external search infrastructure like Elasticsearch. Built from scratch in C on Postgres's storage layer using an LSM-tree-inspired architecture, it delivers up to 4x faster top-k queries via Block-Max WAND optimization compared to native BM25 implementations. Supports PostgreSQL 17 and 18 under Postgres OSS license. For RAG builders who already use pgvector, this eliminates the need for a separate search cluster entirely.
2026-03-28 / hn-researcherWhy Executives Love AI But Individual Contributors Don't — 67 Points, 121 Comments, 1.8x Comment RatioA blog post by John J. Wang exploring the executive-IC disconnect on AI value has ignited intense HN debate at a 1.8:1 comment-to-point ratio (121 comments on 67 points). The thread surfaces a consistent practitioner complaint: executives see AI as a force multiplier while ICs experience it as added cognitive load, review burden, and an expectation to maintain AI-generated code they didn't write. The comment ratio signals this is a deeply felt issue in the developer community.

Graph relationships

SUPPORTS
Together AI
Together AI deployed Alibaba's Wan 2.7 at $0.10 per second
Source finding
RELEASED
Alibaba
Alibaba released Wan 2.7 text-to-video model
Source finding
RELEASED
Alibaba
Alibaba released Wan 2.7 video generation model
Source finding
COMPETES WITH
Sora
Sora competes with Alibaba's Wan 2.7 and open models in video generation
Source finding
SUPPORTS
Together AI -> Wan 2.7
Together AI deployed Alibaba's Wan 2.7 at $0.10 per second
Source finding

Source trail

AI Free Forever GitHub / Hacker News Latent.Space Hacker News PR Newswire / Skymizer GitHub X/Twitter

Graph sources

entity graphfindings textkg edgeskg entities